Disclaimer: Educational content only, not financial advice. Most retail traders lose money; never risk funds you cannot afford to lose entirely.
Backtesting is running a trading rule against historical price data to see how it would have performed — before you put real money behind it.
It sounds obvious. Almost nobody does it properly. Most beginners skip straight from "this pattern looks strong" to a live trade, which is the equivalent of ordering a product you've never seen from a company you've never heard of.
The two kinds of testing — and the order matters#
- Manual backtesting — you scroll through charts (or an offline chart app), apply your rule, and record pretend trades. Slow, honest, and the best teacher of what your rule actually means.
- Automated backtesting — software runs your rule over thousands of bars and spits out statistics. Fast, but dangerously easy to fool yourself with.
Do manual first. If your rule can't survive you personally applying it to 50 trades, no software will save the idea.
How to backtest by hand in an afternoon#
Pick a clean, mechanical rule first. For example: "After a pullback of at least 3 days in an uptrend, buy when price closes back above the 5-day high; stop below the pullback low; target 2× risk."
Now run it like this:
- Pick a market and timeframe you did not just watch. Start 200–300 bars back on a daily chart.
- Scroll bar by bar. Every time the rule triggers, write down: date, entry, stop, target, and outcome (win/loss/R).
- Don't skip rules that lost. Skipping losers while keeping winners is the #1 way to fake your own edge.
- Do at least 30–50 trades before you compute anything.
Then put the rows in a spreadsheet and calculate the three numbers you'll trust: win rate, average win/loss in R, and expectancy per trade. If expectancy isn't clearly positive over 40+ trades, the rule is not tradeable — yet.
The trap every beginner falls into: curve-fitting#
Once you see the data, the temptation is to tweak: "what if the pullback is 2 days instead of 3? What if the target is 3× risk?" You keep adjusting until the historical numbers look beautiful.
That's curve-fitting — building a rule that describes the past perfectly and predicts the future not at all. The fixes:
- Decide the rule first, test after. Every parameter you change while looking at results is a lie you're telling yourself.
- Hold out data. Reserve the last 15–20% of your price history. Never look at it while testing. When you're done, run the rule once on the held-out data, unadjusted. If the edge collapses, you had no edge.
Minimum sample sizes you should respect#
| Statistic | Minimum sample | Why |
|---|---|---|
| Win rate | 30–50 trades | Below this, luck dominates the number |
| Expectancy | 50+ trades | Small-R edges need the long run to appear |
| Consecutive-loss survival | Equal to your worst historical streak +50% | For position sizing, not optimism |
Your plan's stop-loss and position size should assume your worst backtested losing streak happens twice in a row live. If your account dies at that point, the size is too big — see Learn Trading, Part 3: Risk Management.
What backtesting can't tell you#
It can't tell you how you'll behave when the 9th loss in a row arrives — which is why the trading journal and a written trading plan are non-negotiable companions. Data predicts the market's distribution; only discipline predicts you.
Backtesting is not a crystal ball. It's a filter: it separates "I like this idea" from "this idea survives contact with reality." That filter is the cheapest insurance you'll ever buy in this activity — and it costs nothing but an afternoon and your honesty.
Related: Trading Strategy #1: The Trend Pullback and Learn Trading, Part 4: Strategies, Edges, and Backtesting.
